Dilute ensembles of granular matter (so-called granular gases) are nonlinear systems which exhibit fascinating dynamical behavior far from equilibrium, including non-Gaussian distributions of velocities and rotational velocities, clustering, and violation of energy equipartition. In order to understand their dynamic properties, microgravity experiments were performed in suborbital flights and drop tower experiments. Up to now, the experimental images were evaluated mostly manually. Here, we introduce an approach for automatic 3D tracking of positions and orientations of rod-like particles in a dilute ensemble, based on two-view video data analysis. A two-dimensional (2D) localization of particles is performed using a Mask R-CNN neural network trained on a custom data set. The problem of 3D matching of the particles is solved by minimization of the total reprojection error, and finally, particle trajectories are tracked so that ensemble statistics are extracted. Depending on the required accuracy, the software can work fully self-sustainingly or serve as a base for subsequent manual corrections. The approach can be extended to other 3D and 2D particle tracking problems.

颗粒状物质的稀释集合体（所谓的颗粒状气体）是非线性系统，其表现出令人着迷的动态行为，远非平衡状态，包括速度和旋转速度的非高斯分布，聚类以及违反能量均分的情况。为了了解它们的动态特性，在亚轨道飞行和落塔实验中进行了微重力实验。到目前为止，实验图像大部分是通过人工评估的。在这里，我们基于两视图视频数据分析，介绍了一种自动3D跟踪稀疏集合中棒状粒子的位置和方向的方法。使用在自定义数据集上训练的Mask R-CNN神经网络执行粒子的二维（2D）定位。通过最小化总重投影误差来解决粒子的3D匹配问题，最后跟踪粒子的轨迹，以提取整体统计数据。根据所需的精度，该软件可以完全自我维持运行，或作为后续手动更正的基础。该方法可以扩展到其他3D和2D粒子跟踪问题。